Sue Felshin

Integrating Web resources and lexicons into a natural language query system

The START system responds to natural language queries with answers in text, pictures, and other media. STARTs sentence-level natural language parsing relies on a number of mechanisms to help it process the huge, diverse resources available on the World Wide Web. Blitz, a hybrid heuristic- and corpus-based natural language preprocessor enables START to integrate a large and ever-changing lexicon of proper names, by using heuristic rules and precompiled tables of symbols to preprocess various highly regular and fixed expressions into lexical tokens. LaMeTH, a content-based system for extracting information from HTML documents, assists START by providing a uniform method of accessing information on the Web in real time. These mechanisms have considerably improved STARTS ability to analyze real-world sentences and answer queries through expansion of its lexicon and integration of Web resources.

Answering English Questions using Foreign-Language, Semi-Structured Sources

Despite continuing advances in machine translation technology, users who lack familiarity with particular foreign languages have no good way to find information in those languages. In this paper, we present a technical framework and implemented system that answers English questions on the basis of information in foreign-language, semi-structured sources such as websites. This work helps users locate, with high precision, relevant segments of foreign- language information, and then makes use of existing machine translation services to present that information in English. The resulting technology extends an approach embodied in the START information access system and its supporting Omnibase uniform data access system, and it has been applied to several Chinese and Arabic websites.

Gathering knowledge for a question answering system from heterogeneous information sources

Although vast amounts of information are available electronically today, no effective information access mechanism exists to provide humans with convenient information access. A general, open-domain question answering system is a solution to this problem. We propose an architecture for a collaborative question answering system that contains four primary components: an annotations system for storing knowledge, a ternary expression representation of language, a transformational rule system for handling some complexities of language, and a collaborative mechanism by which ordinary users can contribute new knowledge by teaching the system new information. We have developed a initial prototype, called Webnotator, with which to test these ideas.

Learning to Answer Questions from Wikipedia Infoboxes.

A natural language interface to answers on the Web can help us access information more efficiently. We start with an interesting source of information—infoboxes in Wikipedia that summarize factoid knowledge—and develop a comprehensive approach to answering questions with high precision. We first build a system to access data in infoboxes in a structured manner. We use our system to construct a crowdsourced dataset of over 15,000 highquality, diverse questions. With these questions, we train a convolutional neural network model that outperforms models that achieve top results in similar answer selection tasks.

Temporal Grounding Graphs for Language Understanding with Accrued Visual-Linguistic Context

A robots ability to understand or ground natural language instructions is fundamentally tied to its knowledge about the surrounding world. We present an approach to grounding natural language utterances in the context of factual information gathered through natural-language interactions and past visual observations. A probabilistic model estimates, from a natural language utterance, the objects,relations, and actions that the utterance refers to, the objectives for future robotic actions it implies, and generates a plan to execute those actions while updating a state representation to include newly acquired knowledge from the visual-linguistic context. Grounding a command necessitates a representation for past observations and interactions; however, maintaining the full context consisting of all possible observed objects, attributes, spatial relations, actions, etc., over time is intractable. Instead, our model, Temporal Grounding Graphs, maintains a learned state representation for a belief over factual groundings, those derived from natural-language interactions, and lazily infers new groundings from visual observations using the context implied by the utterance. This work significantly expands the range of language that a robot can understand by incorporating factual knowledge and observations of its workspace in its inference about the meaning and grounding of natural-language utterances.